On a summer afternoon in Chicago’s Pilsen neighborhood, the hulking outlines of an oil-fired power station stand in uneasy contrast to nearby playgrounds and residential streets. For years, facilities like this were regarded as transitional relics—kept alive for emergencies, then penciled in for retirement as cleaner power sources expanded. That assumption is now being upended. The rapid rise of artificial intelligence has introduced a new, relentless source of electricity demand, forcing grid operators and utilities to reconsider plants once deemed obsolete.

Across the United States, the infrastructure that supports AI—data centers packed with energy-hungry servers—has altered the calculus of power planning. These facilities draw electricity around the clock, require near-perfect reliability, and concentrate demand in specific regions rather than spreading it evenly across the grid. As a result, so-called “peaker” plants, designed to run only during brief surges in demand, are being pulled back into regular service. Their return reflects not nostalgia for fossil fuel generation, but the stark reality of a power system struggling to keep pace with a sudden and structurally different load.

How AI Computing Reshapes Electricity Demand

Unlike traditional industrial growth, the electricity footprint of AI is both intense and inflexible. Training large-scale machine-learning models requires sustained, high-density power over long periods, while the real-time inference systems that deploy those models must operate continuously. Data centers cannot easily throttle usage during peak hours or accept rolling outages. Their demand behaves more like a baseline industrial load, yet it expands at a speed utilities are unaccustomed to managing.

This surge has exposed a mismatch between modern power planning and present needs. Over the past two decades, U.S. electricity demand grew slowly as efficiency gains offset population growth and economic expansion. Utilities retired older fossil-fueled plants, invested selectively in new gas capacity, and leaned increasingly on renewable generation. AI has reversed that trend almost overnight. In regions hosting dense clusters of data centers, demand growth projections have been rewritten upward, overwhelming reserve margins that once seemed ample.

The challenge is not simply the quantity of power required, but the timing. Renewable energy has expanded rapidly, yet it remains intermittent. Transmission upgrades and new generation projects can take years to permit and build. In the interim, grid operators are turning to assets that already exist and can respond instantly—peaker plants.

Why Peaker Plants Suddenly Make Economic Sense Again

Peaker plants were built for speed, not efficiency. Many are decades old, fueled by oil or natural gas, and designed to start quickly when demand spikes during heatwaves or cold snaps. They sit idle most of the year, earning revenue through capacity payments that compensate owners for being available in emergencies. Under normal conditions, they are expensive to run and emit more pollution per unit of electricity than modern plants.

AI-driven demand has changed those economics. As power markets tighten, prices paid to ensure reliability have surged. Capacity auctions now reflect scarcity rather than surplus, transforming once-marginal plants into profitable assets. Utilities that had planned to retire peakers are reversing course, withdrawing shutdown notices and extending operating lives. In some cases, regulators have stepped in to mandate continued operation, citing risks to grid stability if capacity is withdrawn too quickly.

This revival is less about preference than necessity. Building new power plants or transmission lines takes time, and large-scale battery storage remains costly and limited in duration. Peakers, by contrast, are already connected, permitted, and capable of delivering power within minutes. For grid operators facing the prospect of blackouts, they represent the fastest available solution.

Grid Stress, Market Signals, and Deferred Retirements

The pressure is most visible in regions with concentrated data center development. Large grid operators have warned that demand growth is outstripping the pace at which new generation can be added. Market signals reflect this imbalance: payments to generators for being on standby have soared, and utilities are responding rationally to those incentives.

As a result, retirement schedules for oil-, gas-, and coal-fired plants are being rewritten. Facilities that were expected to close within a few years are now being kept alive indefinitely. Some plants originally designed for continuous operation, later downgraded to peakers, are being reconsidered as semi-regular contributors to the grid. The distinction between emergency backup and everyday supply is becoming blurred.

This dynamic underscores a deeper structural issue. The U.S. grid was optimized for a world of predictable growth and gradual change. AI introduces a shock—rapid, localized, and difficult to forecast—that strains existing planning frameworks. Peaker plants are filling the gap, but they are doing so as a stopgap rather than a long-term solution.

Environmental and Community Consequences of the Revival

The return of peaker plants carries consequences beyond market efficiency. Because many were built decades ago, they often lack modern pollution controls. Their smokestacks are shorter, their emissions more concentrated locally, and their operational profiles dirtier than newer facilities. When run more frequently, they can significantly increase local air pollution.

These plants are disproportionately located in low-income neighborhoods and communities of color, reflecting historical patterns of industrial siting. Residents who have already borne the burden of highways, warehouses, and legacy industrial sites now face the prospect of renewed emissions from facilities they believed were on their way out. For communities that fought hard to close older power plants, the extension of peaker operations feels like a reversal of progress.

The environmental trade-off is stark. On a national scale, peakers contribute a small share of total electricity generation. Locally, however, their impact can be outsized. Increased run times translate directly into higher emissions of sulfur dioxide, nitrogen oxides, and particulate matter—pollutants linked to respiratory and cardiovascular illness. The AI boom, though digital in nature, is producing tangible physical consequences in the air surrounding these neighborhoods.

Why Alternatives Are Not Scaling Fast Enough

Energy experts broadly agree that peaker plants are not an ideal long-term response to rising demand. Expanding transmission capacity could allow electricity to flow from regions with surplus generation to those experiencing shortages. Large-scale batteries could store renewable energy and release it during peak periods, reducing reliance on fossil-fueled backups. Demand management and efficiency measures could smooth consumption patterns.

Yet each alternative faces constraints. Transmission projects are slowed by permitting challenges and local opposition. Battery technology is improving, but most installations are designed for short-duration discharge and cannot yet replace hours or days of continuous backup. Demand response programs work best for flexible loads, whereas data centers prioritize reliability over cost savings.

In this context, peaker plants persist because they solve an immediate problem. They provide certainty in a system grappling with uncertainty. For utilities and grid operators, the risk of insufficient power carries economic and political costs that outweigh concerns about prolonging the life of aging infrastructure.

A Temporary Fix Becoming a Structural Feature

The deeper question is whether the revival of peaker plants is a brief detour or the beginning of a longer-term shift. As AI adoption spreads beyond early hubs and into new regions, demand growth could remain elevated for years. If so, reliance on emergency fossil-fueled capacity may become normalized rather than exceptional.

This possibility complicates the energy transition. Policymakers have promoted a vision of gradual decarbonization supported by renewables, storage, and efficiency. The resurgence of peaker plants suggests that demand-side shocks can derail that trajectory, at least temporarily. It also highlights the lag between digital innovation and physical infrastructure. While software can scale in months, power systems move at a slower pace.

For now, grid operators emphasize pragmatism. Keeping existing plants online is framed as a necessary bridge until new resources come online. Critics counter that repeated extensions risk locking in pollution and delaying investment in cleaner alternatives. The tension between reliability and sustainability is becoming more acute as AI reshapes the energy landscape.

In neighborhoods like Pilsen, that tension is felt not as an abstract policy debate but as a lived reality. The machines powering the future of artificial intelligence are tethered to the remnants of the past, and the cost of that connection is being paid unevenly. As long as AI’s appetite for electricity continues to grow faster than the grid can adapt, the emergency power plants of yesterday will remain an uncomfortable but indispensable part of today’s energy system.

(Adapted from USNews.com)